Hyperparameters

KKTHardNet accepts a training dictionary that is normalized into kkthn.training.KKTTrainConfig.

Common keys

Key	Meaning
`epochs`	Number of training epochs.
`batch_size`	Training batch size.
`learning_rate`	Adam learning rate.
`hidden_size`	Width of hidden MLP layers.
`hidden_layers`	Number of hidden MLP layers.
`train_frac`	Fraction of data used for training.
`seed`	Random seed for splitting and initialization.
`dtype`	Numeric precision, typically `float64`.
`print_every`	Console logging frequency.
`eta`	Optional loss threshold for starting projection training.
`epoch_mlp`	Optional epoch for starting projection training.
`cons_alpha`	Consistency-loss weight.

Projection keys

Key	Meaning
`fb_eps`	Fischer-Burmeister smoothing value.
`gn_max_iters`	Maximum projection solver iterations.
`gn_tol`	Projection residual tolerance.
`gn_reg`	Projection solve regularization.
`newton_step_length`	Initial step length for line search.
`armijo_alpha`	Armijo sufficient-decrease coefficient.
`armijo_beta`	Backtracking contraction coefficient.
`max_backtrack_iter`	Maximum backtracking steps.
`backward_reg`	Regularization in the implicit backward solve.

Core code reference

class kkthn.training.KKTTrainConfig(epochs: 'int' = 1200, batch_size: 'int' = 32, learning_rate: 'float' = 0.001, train_frac: 'float' = 0.8, hidden_size: 'int' = 64, hidden_layers: 'int' = 2, seed: 'int' = 42, dtype: 'str' = 'float64', print_every: 'int' = 1, drop_last: 'bool' = False, eta: 'float | None' = None, epoch_mlp: 'int | None' = None, cons_alpha: 'float' = 0.0, projection: 'ProjectionSettings' = ProjectionSettings(fb_eps=1e-08, gn_max_iters=30, gn_tol=1e-06, gn_reg=0.001, newton_step_length=0.5, armijo_alpha=0.0001, armijo_beta=0.5, max_backtrack_iter=10, armijo_max_steps=10, backward_reg=1e-08))[source]

Bases: object

epochs: int = 1200

batch_size: int = 32

learning_rate: float = 0.001

train_frac: float = 0.8

hidden_size: int = 64

hidden_layers: int = 2

seed: int = 42

dtype: str = 'float64'

print_every: int = 1

drop_last: bool = False

eta: float | None = None

epoch_mlp: int | None = None

cons_alpha: float = 0.0

projection: ProjectionSettings = ProjectionSettings(fb_eps=1e-08, gn_max_iters=30, gn_tol=1e-06, gn_reg=0.001, newton_step_length=0.5, armijo_alpha=0.0001, armijo_beta=0.5, max_backtrack_iter=10, armijo_max_steps=10, backward_reg=1e-08)

class kkthn.projection.ProjectionSettings(fb_eps: 'float' = 1e-08, gn_max_iters: 'int' = 30, gn_tol: 'float' = 1e-06, gn_reg: 'float' = 0.001, newton_step_length: 'float' = 0.5, armijo_alpha: 'float' = 0.0001, armijo_beta: 'float' = 0.5, max_backtrack_iter: 'int' = 10, armijo_max_steps: 'int' = 10, backward_reg: 'float' = 1e-08)[source]

Bases: object

fb_eps: float = 1e-08

gn_max_iters: int = 30

gn_tol: float = 1e-06

gn_reg: float = 0.001

newton_step_length: float = 0.5

armijo_alpha: float = 0.0001

armijo_beta: float = 0.5

max_backtrack_iter: int = 10

armijo_max_steps: int = 10

backward_reg: float = 1e-08